Contrastive Decoding Mitigates Score Range Bias in LLM-as-a-Judge

Large Language Models (LLMs) are commonly used as evaluators in various applications, but the reliability of the outcomes remains a challenge. One such challenge is using LLMs-asjudges for direct assessment, i.e., assigning scores from a specified range without any references. Focusing on summarization, we first show that this challenge stems from LLM judge outputs being associated with score range bias, i.e., LLM judge outputs are highly sensitive to pre-defined score ranges. We also show that similar biases exist among models from the same family. We then mitigate this bias through contrastive decoding, achieving up to 11.7% relative improvement on average in Spearman correlation with human judgments across different score ranges. 1 We stopped at 7 inspired by Likert (1932) showing high correlation between 5 points (1-5) and 7 points (1-7) results. 2 We leave models like Prometheus (Kim et al., 2024) specifically fineutuned on judge tasks as future work since multiple model sizes are not available for contrastive decoding and those models are finetuned towards 1-5 score range. Model Range Pear. Spear. Kend. Llama 3.2-1B 0 to 4 .